1. Introduction

Supercell thunderstorms over eastern North Dakota produced several tornadoes on the evening of 18 July 2004. The smaller of two tornadic supercell storms during this event produced an F4 tornado (Fig. 1). This was only the third F4 tornado documented in North Dakota during the 26-yr period from 1979 to 2004 by the National Oceanic and Atmospheric Administration’s (NOAA’s) publication Storm Data. The environment exhibited large convective available potential energy (CAPE), and significant low-level and deep-layer vertical shear, as is seen in many strong and violent tornado events (e.g., Davies and Johns 1993; Rasmussen and Blanchard 1998; Thompson et al. 2003). However, lifting condensation level (LCL) heights and cloud bases were relatively high in the area where the violent tornadic storm developed, as depicted in the 0000 UTC 19 July 2004 mesoanalysis (Fig. 2) from the Storm Prediction Center (SPC; Bothwell et al. 2002). These latter characteristics and associated surface dewpoint depressions did not appear to be strongly supportive of strong or violent tornado development based on recent statistical studies of supercell tornado environments (e.g., Rasmussen and Blanchard 1998; Markowski et al. 2002; Thompson et al. 2003). Another interesting aspect of this case was the underforecast of surface moisture by operational forecast models. This may have reflected, in part, the inability of the models to properly account for evapotranspiration (ET; Holt et al. 2006). These observations motivated a study of the synoptic and mesoscale environment associated with this event, which is presented in this paper.

The following section is a brief documentation of the tornadoes that occurred on 18 July 2004. An analysis of the associated synoptic and mesoscale environment over eastern North Dakota, including common parameters used in supercell tornado forecasting, will then be presented. The possible effects of ET on boundary layer moisture, CAPE, and convection initiation are considered. The low-level thermodynamic characteristics potentially relevant to tornado development in a relatively high LCL environment are also reviewed. A discussion and summary conclude the paper.

2. Overview of event

There were at least eight distinct tornado reports in the Grand Forks Weather Forecast Office (WFO) county warning area (CWA) during the evening of 18 July 2004. The damage paths and initiation times of these tornadoes, based on two damage survey teams from the Grand Forks WFO, are shown in Fig. 3a. The initial supercell storm developed after 2300 UTC near Grand Forks, and moved south. Several sequential mesocyclones, identified on radar, developed on the southwestern side of this storm between 0000 and 0300 UTC (19 July 2004). These mesocyclones produced several tornadoes, two of which were rated F2 on the Fujita scale and caused around $750,000 in damage. The strongest tornado, rated F4, occurred with a separate supercell in southwestern Barnes County around 0125 UTC, and caused nearly $2 million in damage. The Barnes County tornado appeared to be considerably stronger than the F2 tornadoes farther east when damage was surveyed in both areas at farmsteads that were in the path of the tornadoes. Some of the most significant F4 damage can be seen in Figs. 3b and 3c. Satellite imagery (Fig. 4) shows the two supercells prior to the F4 tornado touchdown at 0125 UTC. The smaller supercell to the southwest, with a pronounced overshooting top, produced the violent tornado.

All tornadoes generally moved from north to south, with an average supercell storm motion of 360° at 9 m s⁻¹. The F4 tornado was approximately 200 yards wide, and had a pathlength of 10 mi. Both F2 tornadoes had pathlengths of 3–4 mi, and were approximately 100 yards wide. No injuries were reported with any of the tornadoes; however, at least 35 cows perished in the F4 tornado. As seen in Figs. 3b and 3c, damage to one farmstead by the F4 tornado was severe. Most structures in the direct path were totally destroyed.

3. Synoptic and mesoscale setting

During the afternoon and evening of 18 July 2004, a ridge axis at 500 hPa (Fig. 5) extended from Colorado to Montana. At 300 hPa (not shown), a 35 m s⁻¹ speed maximum propagated into the region. There was weak upper-level diffluence over southeastern North Dakota around the time the tornadoes occurred. At 850 hPa (not shown), low-level moisture was relatively high over the Dakotas (dewpoints of 17°–20°C), where there was relatively weak westerly flow. Temperatures were +11°C or greater at 700 hPa (not shown) over much of North Dakota.

In the early afternoon at the surface, a pressure trough in central North Dakota moved slowly eastward. By 0100 UTC, the trough was located just southeast of Jamestown, North Dakota (KJMS; Fig. 6). A surface thermal ridge axis, with temperatures in excess of 90°F, extended from near Bismarck, North Dakota (KBIS), to just east of KJMS. An axis of 70°–72°F surface dewpoints was oriented south to north over eastern North Dakota—perpendicular to the thermal ridge. These dewpoints, likely aided by ET (to be discussed later in section 5), enhanced the instability where the highest surface temperatures were collocated with the highest dewpoints. Surface dewpoint depressions increased west of where the tornadic storms developed, with a dewpoint of 15°C at KBIS at 2300 UTC. A very moist boundary layer, and the wind shift near the trough, generated surface moisture flux convergence (not shown), contributing to upward low-level vertical motion over eastern North Dakota (Banacos and Schultz 2005).

4. Environmental characteristics

Instability on the evening of 18 July 2004 was large, and mixed-layer CAPE (MLCAPE) was around 3000 J kg⁻¹ [all mixed-layer computations in this paper use the mean mixing ratio and temperature of the lowest 100 hPa, similar to Thompson et al. (2003)]. The 0000 UTC 19 July 2004 Eta Model analysis in Fig. 7 indicated that MLCAPE was largest close to where the F4 tornadic storm developed. Although mixed-layer convective inhibition (MLCIN, not shown) was also large (−150 to −200 J kg⁻¹) over much of eastern North Dakota through most of the afternoon, heating and lift due to convergence along the surface trough initiated convection by late afternoon. Because convection developed, it is presumed that CIN was reduced sufficiently to allow initiation of several storms, including the violent tornadic supercell that intensified rapidly after 0000 UTC. When solar insolation ceased, all storms weakened quickly, suggesting that this event was driven largely by surface heating and low-level thermodynamics. The 0000 UTC 19 July 2004 observed sounding at Aberdeen, South Dakota (KABR; Fig. 8), confirmed that the environment to the south of the violent tornadic supercell was well capped, with MLCIN near −150 J kg⁻¹.

The 0–6-km shear as depicted on the early evening SPC mesoanalysis Web page was strong over eastern North Dakota (>25 m s⁻¹; Fig. 9) and supportive of supercells (e.g., Rasmussen and Blanchard 1998; Thompson et al. 2003). In addition, the observed hodograph at KABR at 0000 UTC (Fig. 10) was relatively straight above 2 km, similar to the long-lived supercell hodographs in Bunkers et al. (2006). The hodograph was also strongly clockwise curved in the lowest 2 km, which would favor cyclonic rotation of right-moving supercells (Rotunno and Klemp 1982). Based on estimated storm motions, the storm-relative helicity (SRH; Davies-Jones et al. 1990) in the 0–1-km layer from the KABR sounding was near 150 m² s⁻². This is within the range associated with significant tornadoes (e.g., Rasmussen and Blanchard 1998; Thompson et al. 2003). Low-level storm-relative flow (SRF; Kerr and Darkow 1996) was also strong, and from the south (15 m s⁻¹ in the 0–1-km layer). This suggested strong inflow to the supercells over eastern North Dakota from the unstable low-level air mass southeast of the surface trough.

Combinations of CAPE and SRH were large as well. Values of the energy–helicity index (EHI; Hart and Korotky 1991; Davies 1993) from the SPC mesoanalysis Web page at 0000 UTC on 19 July 2004 (Fig. 11) were sizable over southeastern North Dakota (>3.0) based on 0–1-km SRH. This suggested statistical support for significant tornadoes (e.g., Rasmussen 2003; Thompson et al. 2003). Another useful severe weather forecast tool, based on operational experience at NWS Grand Forks, is the vorticity generation parameter (VGP; Rasmussen and Blanchard 1998). The VGP was large (0.4–0.7; not shown) in model analyses, also suggesting support for supercell tornadoes.

One environmental characteristic, the LCL height as depicted in the SPC 0000 UTC 19 July 2004 mesoanalysis (Fig. 2), did not appear to be particularly favorable prior to the development of the F4 tornado. Empirical statistical studies such as Rasmussen and Blanchard (1998), Thompson et al. (2003), and Craven and Brooks (2004) have shown that most significant tornadoes are associated with mixed-layer LCL (MLLCL) heights below 1300–1500 m above ground. However, in this case, MLLCL heights from the SPC mesoanalysis at 0000 UTC (Fig. 2) were rather high, in the 1600–1800-m range. This will be examined later in section 6.

5. Potential effects of ET on low-level moisture and convective initiation

ET frequently has important implications for convection initiation and subsequent storm strength. Raddatz (2000) showed that ET is a secondary (to advected moisture) but significant moisture source for the Canadian prairie provinces. Evapotranspiration also influences the timing and location of convection initiation (Hanesiak et al. 2004). Enhanced moisture, due to ET, raises CAPE values, which in turn increases the severity (Clark and Arritt 1995) of convection. Raddatz (1998) linked vegetative development and the seasonal pattern of ET to the annual pattern of convection. Raddatz and Cummine (2003) linked the peak occurrence of tornado days in the Canadian prairie provinces to the middle of the growing season when ET was greatest. These findings can be applied to the northern plains of the United States, which has a similar agroecosystem. Johns et al. (2000) suggested that ET plays a significant role in strong and violent tornado episodes across this region by increasing CAPE. Observational results (Segal et al. 1989) and numerical simulations (Chang and Wetzel 1991) have also suggested that convergence zones collocated with axes of enhanced boundary layer moisture are influenced by ET. It is important to note that axes of enhanced moisture along convergence zones can only occur when a local moisture source, such as ET, is present.

The middle of the growing season in the northern plains and resultant increase in ET appeared to affect the 18 July 2004 case. Low-level moisture was underforecast by most computer models, leading to an underestimation of convective potential at 0000 UTC 19 July 2004. As an example, the 1200 UTC 18 July 2004 Eta Model run (Fig. 12a) indicated that regional surface dewpoints would rise to around 65°F during the afternoon of 18 July 2004. Instead, dewpoints near the peak temperature time at 0000 UTC 19 July 2004 (Fig. 12b) over eastern North Dakota increased to 70°–73°F ahead of the surface trough. The fact that morning dewpoints (Fig. 12c) were only in the low to mid-60s (°F) over South Dakota, Nebraska, and Kansas with southerly surface winds across that area throughout the day implies that advection did not play a major role in raising dewpoints where the tornadoes occurred. It therefore appears that ET added moisture to the boundary layer during diurnal heating ahead of the surface trough, and was primarily responsible for increasing surface dewpoints over eastern North Dakota beyond levels forecast by operational models.

This enhanced low-level moisture increased CAPE, lowered CIN values, and likely affected convection initiation and intensity. Figure 13 shows an unmodified forecast sounding in Barnes County near the surface trough at 0000 UTC 19 July 2004 from the 1200 UTC Eta Model run. Figure 14 shows the same sounding modified in the lowest 1000 m based on observed surface observations. The lowest 1000 m was randomly chosen for adjustment because this is similar to the lowest 100-hPa layer typically used in MLCAPE computations (e.g., Thompson et al. 2003). This modification increased MLCAPE from 1886 to 3802 J kg⁻¹, and reduced MLCIN from −46 to −1 J kg⁻¹. In addition, the 0–3-km MLCAPE (Rasmussen 2003) increased from 46 to 160 kg⁻¹. This would imply stronger updrafts originating within the boundary layer.

The 1200 UTC 18 July 2004 Eta Model forecast did not generate precipitation over eastern North Dakota. This was probably due in part to the large MLCIN that was forecast. However, with boundary layer moisture increased and MLCIN reduced due to ET, convergence along the trough was sufficient to initiate severe convection. The intensity of the thunderstorms that developed over North Dakota was probably influenced by the increased CAPE, which appeared to be enhanced by ET.

6. LCL height and low-level thermodynamic characteristics

As noted in section 4, MLLCL heights in the Barnes County area (Fig. 2) where the F4 tornado developed appeared somewhat unfavorable (1600–1800 m above ground) for violent tornado development based on statistical research regarding supercell tornado environments. Furthermore, observed surface dewpoint depressions near the Barnes County area were around 10°C, a range suggested to be “nontornadic” by Markowski et al. (2002) when considering storm inflow and rear-flank downdraft characteristics.

A recent study by Davies (2006a) suggests that many “high LCL” tornadic environments combine very steep low-level lapse rates with low-level CAPE along with a level of free convection (LFC) not far above the LCL. Even with relatively high cloud bases (e.g., 1500–2000 m above ground), such a thermodynamic stratification would likely enhance parcel ascent within the lowest portion of the updrafts, with little, if any, low-level CIN present. This might assist tornado development via low-level stretching before significant downdraft cold pools developed that could interfere with surface circulations beneath relatively high-based thunderstorms.

Notice that the model sounding in Fig. 14 that was modified to reflect observed surface moisture had important thermodynamic characteristics. The lapse rate below 2 km was steep (near superadiabatic), and there was significant CAPE below 3 km. In addition, there was little CIN to inhibit rising mixed-layer parcels, and the LFC height was not far above the LCL. These characteristics were also shown by the SPC mesoanalysis over eastern North Dakota in the Jamestown area during the late afternoon of 18 July 2004. A southwest-to-northeast axis of very steep low-level lapse rates was evident by 2300 UTC (Fig. 15), impinging on an area of significant 0–3-km CAPE (Fig. 16). Figure 17 shows the overlap between these two fields 2 h before the tornado formed, suggesting an environment that could enhance low-level stretching beneath cloud bases and within the lower portion of sustained updrafts.

As suggested in section 4, CAPE–shear characteristics over eastern North Dakota were already quite supportive of supercells and tornadoes on the evening of 18 July 2004. As explained in Davies (2006a), the addition of the low-level thermodynamic environment described above may have contributed to intense tornado development with the supercell over Barnes County. A violent tornado occurred even though LCL heights (Fig. 2) appeared noticeably higher than those in the vicinity of the larger-diameter tornadic thunderstorm to the northeast.

7. Summary and conclusions

Apart from LCL height, environmental parameters over North Dakota on the evening of 18 July 2004, such as SRH, deep-layer shear, and CAPE–shear combinations, were quite supportive of supercell tornadoes based on recent research. The strongest tornado of this event, rated F4, formed with a supercell that developed in a high-LCL environment well to the west of a larger supercell that produced weaker tornadoes associated with somewhat lower LCL heights. Although it is not possible to claim a role for environmental variability in governing the intensity of tornadoes on a given day within a small region, it is worth noting that the F4 tornadic thunderstorm formed in an area of maximum surface temperature and dewpoints, resulting in considerable instability. MLCAPE for this event was quite large (around 3000 J kg⁻¹), similar to the findings of Davies (2006b) that showed large values of MLCAPE to be associated with significant tornado cases where LCL heights are relatively high. The fact that low-level lapse rates were also maximized and collocated with relatively large low-level MLCAPE values along the surface trough and wind shift boundary (discussed in section 6) may have had an additional affect on tornado intensity via enhanced stretching, in spite of the high LFC height.

It should be emphasized that prior studies regarding tornadoes and LCL heights have been statistical in nature, and were likely biased toward the eastern United States and lower ground elevations, where higher F-scale ratings are easier to garner due to higher population density and more structures. Similar to the cases in Davies (2006a, b), this event is a reminder that LCL height as a “limiting factor” should be used with caution, especially when other thermodynamic and kinematic factors appear quite favorable for supercells and tornadoes. Future research, including storm modeling and scientific field observations, needs to focus on the physics of the role of low-level stability regarding tornadic supercells, including environments with relatively high LCL heights.

The supercell that eventually produced the F4 tornado was relatively small in size compared to the other tornadic supercell during this event (Fig. 18). The larger supercell appeared more “classic” on radar [larger size, hook-echo “appendage,” outflow boundary, etc. as discussed by Lemon and Doswell (1979)] than the smaller, younger supercell to the southwest, yet it produced weaker tornadoes. This reaffirms that there is not necessarily an association between the radar size/appearance of a supercell and the intensity of tornadoes produced. This case demonstrates that a relatively small and newly developed supercell can produce strong or violent tornadoes if the local environment is favorable.

The ET, discussed in section 5, likely had an impact on the storm environment. Moisture added to the boundary layer ahead of the surface trough by ET appeared to be responsible for enhanced surface dewpoints over eastern North Dakota that were not forecast by operational models. Because this added moisture likely affected convective initiation and intensity, this case appears to emphasize the importance of ET as a factor that forecasters need to consider when making thunderstorm forecasts during the growing season.

Presenting several operational challenges, the authors believe this case serves as an example of several localized factors coming together to generate a violent tornado. The recognition of certain environmental variables, such as enhanced moisture and surface heating along a slow-moving boundary near the periphery of a capping inversion, should suggest heightened situational awareness for severe weather forecasters. In such cases, though LCL heights may appear relatively high in a statistical sense (e.g., 1500–2000 m), favorable CAPE and shear parameters in combination with the presence of steep low-level lapse rates may signal the potential for significant supercell tornado development.